Presentation Schedule
Adaptive Tactical Decision Making in Ice Hockey: Integrating Multi-Agent Reinforcement Learning Framework with Advanced Computer Vision Techniques (93889)
Session: On Demand
Room: Virtual Video Presentation
Presentation Type:Virtual Presentation
This study proposes an integrated framework that combines with Multi Agent Reinforcement Learning (MARL) with advanced computer vision techniques, including pose recognition via MMaction2, object detection with Yolov11, and multi-object tracking using ByteTrack. The McGill Hockey Player Tracking Dataset (MHPTD) is used to analysis the ice hockey team sport. In building the framework with MARL, each player is modeled as an autonomous agent whose observation space encompasses self-position, puck state, and the spatial locations of teammates and opponents. By incorporating inputs from the MHPTD dataset, the framework dynamically adapts strategic behaviors, like defending, attacking. When simulating realistic ice hockey scenarios, the framework can be used for strategy optimization using the group object detection, play behavior modeling by pose recognition and positional data tracking, advanced AI opponents’ strategic analysis in future games. Reward function is setup to encourage agents to move towards the puck and getting rewards when shooting to the opponents’ door. Multi-agent setup can simulate full team sport. The framework can enhance the simulation of complex player interactions, bridge the gap between MARL simulation and real-world fix-field team sports, provide insight in coaching and sport education.
Authors:
Boyang Zhang, University of Turku, Finland
About the Presenter(s)
Senior Post-doc researcher in University of Turku, Finland
See this presentation on the full schedule – On Demand Schedule
Virtual Presentation
Conference Comments & Feedback
Place a comment using your LinkedIn profileShare this Presentation
Comments
Powered by WP LinkPress